Lecture 4: Data Visualization

BIOF 339

October 3, 2018

Data Visualization in R

ggplot2

We’re making the decision to use ggplot2 for graphics

  • Makes pretty good formatting choices out of the box
  • Works like pipes!!
  • Is declarative (tell it what you want) without getting caught up in minutae
  • Strongly leverages data frames (good practice)
  • Fast enough
  • There are good templates if you want to change the look

Introduction to ggplot2

# install.packages('tidyverse')
library(ggplot2)

Introduction to ggplot2

The ggplot2 package is a very flexible and (to me) intuitive way of visualizing data. It is based on the concept of layering elements on a canvas.

This idea of layering graphics on a canvas is, to me, a nice way of building graphs

Introduction to ggplot2

You need:

  • A data.frame object
  • Aesthetic mappings (aes) to say what data is used for what purpose in the viz
    • x- and y-direction
    • shapes, colors, lines
  • A geometry object (geom) to say what to draw
    • You can “layer” geoms on each other to build plots

Introduction to ggplot2

ggplot used pipes before pipes were a thing.

However, it uses the + symbol for piping rather than the %>% operator, since it pre-dates the tidyverse

Introduction to ggplot2

library(ggplot2)
ggplot(mtcars, aes(x = wt, y = mpg)) + geom_point()
  • A data.frame object: mtcars
  • Aesthetic mapping:
    • x-axis: wt
    • y-axis: mpg
  • Geometry:
    • geom_point: draw points

Introduction to ggplot2

library(ggplot2)
ggplot(mtcars, aes(x = wt, y = mpg)) + geom_point()+ geom_smooth()
  • A data.frame object: mtcars
  • Aesthetic mapping:
    • x-axis: wt
    • y-axis: mpg
  • Geometry:
    • geom_point: draw points
    • geom_smooth: Add a layer which draws a best-fitting line

ggplot2 examples

We will use the two data sets:

data_spine <- read.csv('http://www.araastat.com/BIOF339_PracticalR/
                       Lectures/lecturedataframe_data/Dataset_spine.csv', 
                       stringsAsFactors = F)

data_brca <- read.csv('http://www.araastat.com/BIOF339_PracticalR/
                      Lectures/lecturedataframe_data/
                      clinical_data_breast_cancer_modified.csv',
                      stringsAsFactors = F)

Plotting one variable

Histograms

ggplot(data_brca, aes(x = Age.at.Initial.Pathologic.Diagnosis)) + 
  geom_histogram()

Histograms

ggplot(data_brca, aes(x = Age.at.Initial.Pathologic.Diagnosis)) + 
  geom_histogram(binwidth=4)

Density plot

ggplot(data_brca, aes(x = Age.at.Initial.Pathologic.Diagnosis)) + 
  geom_density()

Bar plot

ggplot(data_brca, aes(x = Tumor))+geom_bar()

Exercise

Exercise

Using the mtcars dataset in R, create:

  1. A histogram of the fuel efficiences (mpg) in the data set
  2. A bar plot of frequencies of number of cylinders (cyl) in the car

Solution

ggplot(mtcars, aes(x = mpg)) + geom_histogram(binwidth=3)

# ggplot(mtcars) + geom_histogram(aes(x = mpg), binwidth = 3)

Solution

ggplot(mtcars, aes(x = factor(cyl))) + geom_bar()

Two continuous variables

Scatter plots

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope)) + 
  geom_point()

Scatter plot with a smooth line

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope))+
  geom_point() + 
  geom_smooth()

Scatter plot with a smooth straight line

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope)) +
  geom_point()+
  geom_smooth(method='lm')

Line plot (for time series)

library(forecast)
d <- data.frame(x = 1:length(gas), y = gas) # Australian monthly gas production
ggplot(d, aes(x, y)) + geom_line()

Exercise

Exercise

  1. Create a scatter plot of sepal length and sepal width from the iris dataset, and add a smooth line through it

Solution

ggplot(iris, aes(Sepal.Length, Sepal.Width)) + geom_point() + geom_smooth()

Continuous variable with discrete variable

Boxplots

ggplot(data_spine, aes(x = Class.attribute, y = Sacral.slope))+
  geom_boxplot()

# Factor/discrete variable is always x

Violin plots

ggplot(data_spine, aes(x = Class.attribute, y = Sacral.slope)) +
  geom_violin()

Exercise

Exercise

  1. Plot a boxplot of petal length by species using the iris dataset

Solution

ggplot(iris, aes(x = Species, y = Petal.Length))+geom_boxplot()

Flipping axes

Vertical bars

ggplot(data_brca, aes(x = Tumor))+geom_bar()

Horizontal bars

ggplot(data_brca, aes(x = Tumor))+geom_bar()+
  coord_flip()

Resources

Online resources

  • The ggplot website has many resources to help create visualizations
  • The R Graph Gallery
  • There are a lot of blogs showing many capabilities of ggplot2
  • StackOverflow is the place for Q & A.

Other packages

  • The cowplot and ggpubr packages provide several improvements on ggplot2, including more themes and an easy way to put several graphs together in a panel

  • There is a rich ecosystem around ggplot2, which can be explored here

Group-wise descriptives and visualizations

Grouping

  • It is common to look at statistics within subgroups of the data
  • The idea is to see if secondary variables affect your primary outcome or relationship

Introducing the dplyr package

dplyr is the most lucid package for manipulating and analyzing data organized in a data frame.

  • It has a group_by function which creates a grouped data frame
library(dplyr)
grouped_data_spine = data_spine %>% group_by(Class.attribute)

Note that you have to group using a discrete valued variable (factor, character, integer)

Grouped summaries

grouped_data_spine %>% 
  summarize(mean(Pelvic.incidence), 
          sd(Pelvic.incidence),
          min(Pelvic.incidence),
          max(Pelvic.incidence))
Class.attribute mean(Pelvic.incidence) sd(Pelvic.incidence) min(Pelvic.incidence) max(Pelvic.incidence)
Abnormal 64.69 17.66 26.15 129.83
Normal 51.69 12.37 30.74 89.83

Grouped summaries

grouped_data_spine %>% summarize(Mean = mean(Pelvic.incidence), 
                                 SD = sd(Pelvic.incidence),
                                 Min = min(Pelvic.incidence),
                                 Max = max(Pelvic.incidence))
Class.attribute Mean SD Min Max —————- —— —— —— ——- Abnormal 64.69 17.66 26.15 129.83 Normal 51.69 12.37 30.74 89.83

Grouped summaries

grouped_data_spine %>% summarize_all(mean)
#   # A tibble: 2 x 13
#     Class.attribute Pelvic.incidence Pelvic.tilt Lumbar.lordosis…
#     <fct>                      <dbl>       <dbl>            <dbl>
#   1 Abnormal                    64.7        19.8             55.9
#   2 Normal                      51.7        12.8             43.5
#   # ... with 9 more variables: Sacral.slope <dbl>, Pelvic.radius <dbl>,
#   #   Degree.spondylolisthesis <dbl>, Pelvic.slope <dbl>, Direct.tilt <dbl>,
#   #   Thoracic.slope <dbl>, Cervical.tilt <dbl>, Sacrum.angle <dbl>,
#   #   Scoliosis.slope <dbl>

A note on tibbles

  • Tibbles are a new-generation object meant to enhance the data.frame.
  • If you want to just get back to a more familiar data.frame object, use as.data.frame
  • A tibble is built on a data.frame, so all operations on data.frame’s will work.
  • To see all columns, set options(dplyr.width=Inf).

A note on tibbles

Differences between a tibble and a data.frame:

  1. Printing a tibble is restricted to the first 10 lines, and includes column types
  2. Stricter subsetting rules that make the types of objects created consistent

Using ggplot in a pipeline

data_spine %>% 
  group_by(Class.attribute) %>% 
  summarize_all(funs(Mean = mean(., na.rm=T),
                     SEM = sd(., na.rm=T)/sqrt(n()))) %>% 
  gather(variable, value, -Class.attribute) %>% 
  separate(variable, c('Variable','stat'), sep = '_') %>% 
  spread(stat, value) %>% 
  mutate(lcb = Mean - 2 * SEM, ucb = Mean + 2 * SEM) %>% 
  ggplot(aes(x = Class.attribute, y = Mean, ymin = lcb, ymax = ucb)) +
    geom_pointrange() +
    facet_wrap(~Variable, scales = 'free_y') + 
    labs( x = 'Class', y = '') + 
    ggtitle('Confidence intervals of the mean')

Work through the pipeline yourself to understand what each step does, just like last week

Grouped visualization

Density plot

ggplot(data_spine, aes(x = Sacral.slope, group = Class.attribute, 
                       color=Class.attribute))+
  geom_density()

Scatter plot

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope,
                       group = Class.attribute, color = Class.attribute))+
  geom_point()

Scatter plot (Black and White)

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope, 
                       group = Class.attribute, shape = Class.attribute))+
  geom_point()

Scatter plot with size representing third variable

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope))+
  geom_point(aes(size = Pelvic.slope))

Scatter plot combinations

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope, 
                       group = Class.attribute, color = Class.attribute))+
  geom_point(aes(size = Pelvic.slope))

Scatter plot with lines

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope, 
                       group = Class.attribute, color=Class.attribute))+
  geom_point()+
  geom_smooth(method='lm')

Scatter plot with lines

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope))+
  geom_point()+
  geom_smooth(aes(color = Class.attribute), method='lm')

Facetting

Facetting

Facetted graphs are a panel of graphs, each of which corresponds to a particular subgroup of the data.

Facetted scatter plot

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope))+
  geom_point()+
  facet_wrap( ~ Class.attribute, nrow=1)

Facetted scatter plot with lines

ggplot(data_spine, aes(x = Lumbar.lordosis.angle, y = Sacral.slope))+
  geom_point()+ geom_smooth(method='lm')+
  facet_wrap( ~ Class.attribute, nrow=1)

Manhattan plot

Manhattan plot

library(qqman)
data(gwasResults)
head(gwasResults)
#     SNP CHR BP         P
#   1 rs1   1  1 0.9148060
#   2 rs2   1  2 0.9370754
#   3 rs3   1  3 0.2861395
#   4 rs4   1  4 0.8304476
#   5 rs5   1  5 0.6417455
#   6 rs6   1  6 0.5190959
gwasResults <- gwasResults %>% 
  mutate(x_position = 1:n())

Manhattan plot

ggplot(gwasResults, aes(x = x_position, y = -log(P, base=10)))+
  geom_point(size = 0.2)

Manhattan plot

ggplot(gwasResults, aes(x = x_position, y = -log(P, base=10),
                        group=CHR, color=CHR))+
  geom_point(size=0.2)

Manhattan plot

ggplot(gwasResults, aes(x = x_position, y = -log(P, base=10),
                        group=factor(CHR), color=factor(CHR)))+
  geom_point(size=0.2)

Manhattan plot

ggplot(gwasResults, aes(x = x_position, y = -log(P, base=10),
                        group=factor(CHR), color=factor(CHR)))+
  geom_point(size=0.2)+
  geom_hline(yintercept = 8, color='red', linetype=2)

Manhattan plot, exploded

ggplot(gwasResults, aes(x = BP, y = -log(P, base=10)))+
  geom_point(size=0.2)+
  facet_wrap(~ CHR, nrow=4)+
  geom_hline(yintercept = 8, color='red', linetype=2)

Manhattan plot, exploded